Mercurial > libavcodec.hg
view ps2/idct_mmi.c @ 11305:1a349d669184 libavcodec
Optimize (amvd>2)+(amvd>32), about 1 cpu cycles faster.
patch by Zhou Zongyi @ zhouzy () os punkt pku dot edu speck cn
| author | michael |
|---|---|
| date | Fri, 26 Feb 2010 22:45:35 +0000 |
| parents | eebc7209c47f |
| children |
line wrap: on
line source
/* * Originally provided by Intel at Application Note AP-922. * * Column code adapted from Peter Gubanov. * Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru> * http://www.elecard.com/peter/idct.shtml * rounding trick copyright (c) 2000 Michel Lespinasse <walken@zoy.org> * * MMI port and (c) 2002 by Leon van Stuivenberg * * This file is part of FFmpeg. * * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * FFmpeg is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ #include "libavutil/common.h" #include "libavcodec/dsputil.h" #include "mmi.h" #define BITS_INV_ACC 5 // 4 or 5 for IEEE #define SHIFT_INV_ROW (16 - BITS_INV_ACC) #define SHIFT_INV_COL (1 + BITS_INV_ACC) #define TG1 6518 #define TG2 13573 #define TG3 21895 #define CS4 23170 #define ROUNDER_0 0 #define ROUNDER_1 16 #define TAB_i_04 (32+0) #define TAB_i_17 (32+64) #define TAB_i_26 (32+128) #define TAB_i_35 (32+192) #define TG_1_16 (32+256+0) #define TG_2_16 (32+256+16) #define TG_3_16 (32+256+32) #define COS_4_16 (32+256+48) #define CLIPMAX (32+256+64+0) static short consttable[] align16 = { /* rounder 0*/ // assume SHIFT_INV_ROW == 11 0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1, /* rounder 1*/ 0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0, /* row 0/4*/ 16384, 21407, -16384, -21407, 22725, 19266, -22725, -12873, 8867, 16384, 8867, 16384, 4520, 12873, -4520, 19266, 16384, -8867, 16384, -8867, 12873, -22725, 19266, -22725, 21407, -16384, -21407, 16384, 19266, 4520, -12873, 4520, /* row 1/7*/ 22725, 29692, -22725, -29692, 31521, 26722, -31521, -17855, 12299, 22725, 12299, 22725, 6270, 17855, -6270, 26722, 22725, -12299, 22725, -12299, 17855, -31521, 26722, -31521, 29692, -22725, -29692, 22725, 26722, 6270, -17855, 6270, /* row 2/6*/ 21407, 27969, -21407, -27969, 29692, 25172, -29692, -16819, 11585, 21407, 11585, 21407, 5906, 16819, -5906, 25172, 21407, -11585, 21407, -11585, 16819, -29692, 25172, -29692, 27969, -21407, -27969, 21407, 25172, 5906, -16819, 5906, /*row 3/5*/ 19266, 25172, -19266, -25172, 26722, 22654, -26722, -15137, 10426, 19266, 10426, 19266, 5315, 15137, -5315, 22654, 19266, -10426, 19266, -10426, 15137, -26722, 22654, -26722, 25172, -19266, -25172, 19266, 22654, 5315, -15137, 5315, /*column constants*/ TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3, CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4, /* clamp */ 255, 255, 255, 255, 255, 255, 255, 255 }; #define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \ lq(blk, rowoff, $16); /* r16 = x7 x5 x3 x1 x6 x4 x2 x0 */ \ /*slot*/ \ lq($24, 0+taboff, $17); /* r17 = w */ \ /*delay slot $16*/ \ lq($24, 16+taboff, $18);/* r18 = w */ \ prevh($16, $2); /* r2 = x1 x3 x5 x7 x0 x2 x4 x6 */ \ lq($24, 32+taboff, $19);/* r19 = w */ \ phmadh($17, $16, $17); /* r17 = b1"b0'a1"a0' */ \ lq($24, 48+taboff, $20);/* r20 = w */ \ phmadh($18, $2, $18); /* r18 = b1'b0"a1'a0" */ \ phmadh($19, $16, $19); /* r19 = b3"b2'a3"a2' */ \ phmadh($20, $2, $20); /* r20 = b3'b2"a3'a2" */ \ paddw($17, $18, $17); /* r17 = (b1)(b0)(a1)(a0) */ \ paddw($19, $20, $19); /* r19 = (b3)(b2)(a3)(a2) */ \ pcpyld($19, $17, $18); /* r18 = (a3)(a2)(a1)(a0) */ \ pcpyud($17, $19, $20); /* r20 = (b3)(b2)(b1)(b0) */ \ paddw($18, rnd, $18); /* r18 = (a3)(a2)(a1)(a0) */\ paddw($18, $20, $17); /* r17 = ()()()(a0+b0) */ \ psubw($18, $20, $20); /* r20 = ()()()(a0-b0) */ \ psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \ psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \ ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0 Note order */ \ \ prevh(outreg, $2); \ pcpyud($2, $2, $2); \ pcpyld($2, outreg, outreg); \ } #define DCT_8_INV_COL8() \ \ lq($24, TG_3_16, $2); /* r2 = tn3 */ \ \ pmulth($11, $2, $17); /* r17 = x3 * tn3 (6420) */ \ psraw($17, 15, $17); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \ psubh($17, $13, $17); /* r17 = tm35 */ \ \ pmulth($13, $2, $18); /* r18 = x5 * tn3 (6420) */ \ psraw($18, 15, $18); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $18, $18); /* r18 = x5 * tn3 */ \ paddh($18, $11, $18); /* r18 = tp35 */ \ \ lq($24, TG_1_16, $2); /* r2 = tn1 */ \ \ pmulth($15, $2, $19); /* r19 = x7 * tn1 (6420) */ \ psraw($19, 15, $19); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $19, $19); /* r19 = x7 * tn1 */ \ paddh($19, $9, $19); /* r19 = tp17 */ \ \ pmulth($9, $2, $20); /* r20 = x1 * tn1 (6420) */ \ psraw($20, 15, $20); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $20, $20); /* r20 = x1 * tn1 */ \ psubh($20, $15, $20); /* r20 = tm17 */ \ \ psubh($19, $18, $3); /* r3 = t1 */ \ paddh($20, $17, $16); /* r16 = t2 */ \ psubh($20, $17, $23); /* r23 = b3 */ \ paddh($19, $18, $20); /* r20 = b0 */ \ \ lq($24, COS_4_16, $2); /* r2 = cs4 */ \ \ paddh($3, $16, $21); /* r21 = t1+t2 */ \ psubh($3, $16, $22); /* r22 = t1-t2 */ \ \ pmulth($21, $2, $21); /* r21 = cs4 * (t1+t2) 6420 */ \ psraw($21, 15, $21); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $21, $21); /* r21 = b1 */ \ \ pmulth($22, $2, $22); /* r22 = cs4 * (t1-t2) 6420 */ \ psraw($22, 15, $22); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $22, $22); /* r22 = b2 */ \ \ lq($24, TG_2_16, $2); /* r2 = tn2 */ \ \ pmulth($10, $2, $17); /* r17 = x2 * tn2 (6420) */ \ psraw($17, 15, $17); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \ psubh($17, $14, $17); /* r17 = tm26 */ \ \ pmulth($14, $2, $18); /* r18 = x6 * tn2 (6420) */ \ psraw($18, 15, $18); \ pmfhl_uw($3); /* r3 = 7531 */ \ psraw($3, 15, $3); \ pinteh($3, $18, $18); /* r18 = x6 * tn2 */ \ paddh($18, $10, $18); /* r18 = tp26 */ \ \ paddh($8, $12, $2); /* r2 = tp04 */ \ psubh($8, $12, $3); /* r3 = tm04 */ \ \ paddh($2, $18, $16); /* r16 = a0 */ \ psubh($2, $18, $19); /* r19 = a3 */ \ psubh($3, $17, $18); /* r18 = a2 */ \ paddh($3, $17, $17); /* r17 = a1 */ #define DCT_8_INV_COL8_STORE(blk) \ \ paddh($16, $20, $2); /* y0 a0+b0 */ \ psubh($16, $20, $16); /* y7 a0-b0 */ \ psrah($2, SHIFT_INV_COL, $2); \ psrah($16, SHIFT_INV_COL, $16); \ sq($2, 0, blk); \ sq($16, 112, blk); \ \ paddh($17, $21, $3); /* y1 a1+b1 */ \ psubh($17, $21, $17); /* y6 a1-b1 */ \ psrah($3, SHIFT_INV_COL, $3); \ psrah($17, SHIFT_INV_COL, $17); \ sq($3, 16, blk); \ sq($17, 96, blk); \ \ paddh($18, $22, $2); /* y2 a2+b2 */ \ psubh($18, $22, $18); /* y5 a2-b2 */ \ psrah($2, SHIFT_INV_COL, $2); \ psrah($18, SHIFT_INV_COL, $18); \ sq($2, 32, blk); \ sq($18, 80, blk); \ \ paddh($19, $23, $3); /* y3 a3+b3 */ \ psubh($19, $23, $19); /* y4 a3-b3 */ \ psrah($3, SHIFT_INV_COL, $3); \ psrah($19, SHIFT_INV_COL, $19); \ sq($3, 48, blk); \ sq($19, 64, blk); #define DCT_8_INV_COL8_PMS() \ paddh($16, $20, $2); /* y0 a0+b0 */ \ psubh($16, $20, $20); /* y7 a0-b0 */ \ psrah($2, SHIFT_INV_COL, $16); \ psrah($20, SHIFT_INV_COL, $20); \ \ paddh($17, $21, $3); /* y1 a1+b1 */ \ psubh($17, $21, $21); /* y6 a1-b1 */ \ psrah($3, SHIFT_INV_COL, $17); \ psrah($21, SHIFT_INV_COL, $21); \ \ paddh($18, $22, $2); /* y2 a2+b2 */ \ psubh($18, $22, $22); /* y5 a2-b2 */ \ psrah($2, SHIFT_INV_COL, $18); \ psrah($22, SHIFT_INV_COL, $22); \ \ paddh($19, $23, $3); /* y3 a3+b3 */ \ psubh($19, $23, $23); /* y4 a3-b3 */ \ psrah($3, SHIFT_INV_COL, $19); \ psrah($23, SHIFT_INV_COL, $23); #define PUT(rs) \ pminh(rs, $11, $2); \ pmaxh($2, $0, $2); \ ppacb($0, $2, $2); \ sd3(2, 0, 4); \ __asm__ volatile ("add $4, $5, $4"); #define DCT_8_INV_COL8_PUT() \ PUT($16); \ PUT($17); \ PUT($18); \ PUT($19); \ PUT($23); \ PUT($22); \ PUT($21); \ PUT($20); #define ADD(rs) \ ld3(4, 0, 2); \ pextlb($0, $2, $2); \ paddh($2, rs, $2); \ pminh($2, $11, $2); \ pmaxh($2, $0, $2); \ ppacb($0, $2, $2); \ sd3(2, 0, 4); \ __asm__ volatile ("add $4, $5, $4"); /*fixme: schedule*/ #define DCT_8_INV_COL8_ADD() \ ADD($16); \ ADD($17); \ ADD($18); \ ADD($19); \ ADD($23); \ ADD($22); \ ADD($21); \ ADD($20); void ff_mmi_idct(int16_t * block) { /* $4 = block */ __asm__ volatile("la $24, %0"::"m"(consttable[0])); lq($24, ROUNDER_0, $8); lq($24, ROUNDER_1, $7); DCT_8_INV_ROW1($4, 0, TAB_i_04, $8, $8); DCT_8_INV_ROW1($4, 16, TAB_i_17, $7, $9); DCT_8_INV_ROW1($4, 32, TAB_i_26, $7, $10); DCT_8_INV_ROW1($4, 48, TAB_i_35, $7, $11); DCT_8_INV_ROW1($4, 64, TAB_i_04, $7, $12); DCT_8_INV_ROW1($4, 80, TAB_i_35, $7, $13); DCT_8_INV_ROW1($4, 96, TAB_i_26, $7, $14); DCT_8_INV_ROW1($4, 112, TAB_i_17, $7, $15); DCT_8_INV_COL8(); DCT_8_INV_COL8_STORE($4); //let savedtemp regs be saved __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23"); } void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block) { /* $4 = dest, $5 = line_size, $6 = block */ __asm__ volatile("la $24, %0"::"m"(consttable[0])); lq($24, ROUNDER_0, $8); lq($24, ROUNDER_1, $7); DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8); DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9); DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10); DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11); DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12); DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13); DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14); DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15); DCT_8_INV_COL8(); lq($24, CLIPMAX, $11); DCT_8_INV_COL8_PMS(); DCT_8_INV_COL8_PUT(); //let savedtemp regs be saved __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23"); } void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block) { /* $4 = dest, $5 = line_size, $6 = block */ __asm__ volatile("la $24, %0"::"m"(consttable[0])); lq($24, ROUNDER_0, $8); lq($24, ROUNDER_1, $7); DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8); DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9); DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10); DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11); DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12); DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13); DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14); DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15); DCT_8_INV_COL8(); lq($24, CLIPMAX, $11); DCT_8_INV_COL8_PMS(); DCT_8_INV_COL8_ADD(); //let savedtemp regs be saved __asm__ volatile(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23"); }